From: groepaz (groepaz_at_gmx.net)
Date: 2002-07-13 19:53:18
Hello Ullrich,
UvB> There are many command line tools out there that process the command line from
UvB> left to right.
[snip]
UvB> I hope this explains why things are as they are :-)
oh yeah, well... i understood that before i think :=P i still it'd be
better to make this act somewhat "smarter".... i understand that it
makes sence for certain switches to be used more than once on the
command line (and thus affecting a number of files only, depending on
their order) but IMHO for some others its pretty uncommon atleast. for
eg one could assume only few ppl use cl65 for compiling to multiple output
files, or multiple targets....well i dont know :o) atleast for
cc65 and ca65 (when called manually) it would be nice if they would
work that way...
however, no matter what, it would be smart to be able to be warned
about this possible problem ;)
>> ok, for the -Oi bug its your turn now i guess :o)
UvB> The base routine for printf() contains a
UvB> jmp (ind)
UvB> command. The linker should warn if the argument address crosses a page
UvB> boundary, but it does not. Your code seems to trigger that bug. Please try to
UvB> reorder the array program (move the f() and g() functions before main() in the
UvB> source) and recompile it with the command line switches that were not working.
UvB> If the program works now, it is the bug described above. I have already fixed
UvB> the problem in the development version (using jmp (ind) is never a good idea,
UvB> even with a linker extension, because there are output formats - o65 for
UvB> example - where the final address is not known).
mh wouldnt that be fixed as easy as
.align 2
ind: .word $beaf
? :o)
however, i have rebuilt the compiler and the libs from yesterdays
snapshot.... looks like that cured some more problems :o)
anyway, theres one thing that really bothered me last nite (until i
have found a workaround ;=D), which is appearently broken variable
argument lists in C-funktions...... to be more precise, referring to
the parameters before the variable list by name does not work as
expected.
(i am compiling with no optimizations at again by now... until i have
eliminated all compiler-caused problems that is ;=P)
this code wont work:
void chk0(char *format,...){
va_list ap;
static char string[0x100];
va_start(ap,format);
vsprintf(string,format,ap);
printf("format:%s,string:%s\n",_format,string);
va_end(ap);
}
and this is my current workaround:
void chk0(char *format,...){
__asm__ ("pha"); // save argument size
{
//va_list ap;
char *ap;
char *_format;
static char string[0x100];
// va_start(ap,format);
__asm__ ("pla"); // restore argument size
__asm__ ("ldx #$00"); // clear hibyte of AX
ap=__AX__;
ap+=(char*)&format; // adress of format incorrectly
// equals the adress of the last
// parameter in the variable list
// get value of format
ap-=2;
_format=*((char**)ap);
// vsprintf(string,format,ap);
vsprintf(&string[0],_format,ap);
printf("format:%s,string:%s\n",_format,string);
// va_end(ap);
}
}
tjam... soo :o) i got most of the .c code i made for/with 2.7.1
working again now, i am gonna take a closer look at the filestuff atm
and also rechecking (and possibly removing) some workarounds that were
previously needed. after all that is done i am going to create a
little textfile that documents some of that stuff and which references
to the programs that can be used to trigger the specific bug (or,
which need workarounds).... and finally upload a new snapshot then :o)
after that, i'd like to check out that loadable modules stuff, and
your gfx interface. (btw does there exist a little testprogram or sth
that shows how to use these loadable modules?) ... maybe i can
contribute most of the gfx-related stuff i have done so far.
--
Best regards,
groepaz mailto:groepaz_at_gmx.net
----------------------------------------------------------------------
To unsubscribe from the list send mail to majordomo_at_musoftware.de with
the string "unsubscribe cc65" in the body(!) of the mail.
This archive was generated by hypermail 2.1.3 : 2002-07-13 19:53:59 CEST